Evaluation of approaches to identifying hail damage to crop vegetation using satellite imagery

Abstract

During the growing season in the central United States, severe thunderstorms frequently occur and produce large hail that damages the underlying vegetation, often in agricultural areas. Satellite remote sensing provides a tool for identifying these damaged areas. Previous studies have used changes in the normalized difference vegetation index (NDVI) to identify and examine these areas of damage, but have done so in a manual, time-consuming manner. This study examines an automated approach to detecting areas of hail damage in satellite imagery. Two techniques are evaluated: (i) use of an NDVI change threshold and (ii) detection of anomalies that occur in both daily NDVI and land surface temperature imagery. The two techniques are scored against one another using three different case studies. Two of the case studies occurred late in the growing season in August, and the third occurred in the growing season in early June. The NDVI threshold performed well in the two August case studies with a final probability of detection (POD) ranging from 0.497 to 0.647, whereas the anomaly detection for these two case studies had a lower POD of 0.317 to 0.587. The early June case study highlighted the limitations of using an NDVI threshold and the strengths of using anomaly detection. The POD for the NDVI threshold technique was 0.07–0.08 with a false alarm ratio (FAR) of 0.661–0.758, whereas the anomaly detection had a POD of 0.399–0.418 and a FAR of 0.540–0.681 for this third case study.